Microsoft AI: MAI-Image-2.5 Pro

MAI-Image-2.5 Pro is an image generation model from Microsoft, listed under the model id microsoft/mai-image-2.5-pro. It sits in the image category and is described by the provider as a high-quality image generation model available via Azure AI Foundry. If you are already working inside Microsoft's ecosystem, or you are comparing image models across providers, this is the Microsoft entry to consider.

The model takes text and image inputs and returns images. That means you can drive it with a written prompt, and you can also supply a single reference image alongside that prompt. The context window is 4096 tokens, which covers your prompt and any reference material you pass in. Output is capped at 1024, and generation is limited to one image per request, so each call produces a single result rather than a batch. Aspect ratio is selectable from a fixed set: 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, or auto, which lets you match the output shape to the slot it will fill rather than cropping after the fact.

Those capabilities point to a fairly specific set of jobs. Text to image work covers the usual ground: concept art, marketing visuals, illustrations, and photorealistic renders from a written description. The image input path is what makes the model more interesting for iterative work. You can pass a reference image and a prompt to steer a variation, restyle an existing asset, or explore directions from a starting point instead of a blank page. Because only one reference image is accepted, this suits single anchor workflows rather than multi image composition or collage style tasks. The fixed aspect ratios help when you need output that drops straight into a layout, a storyboard frame, or a social format without resizing.

Calling it through api14 works the same way as any other model on the gateway, using the OpenAI compatible API with the model id microsoft/mai-image-2.5-pro. The supported request parameters are max_completion_tokens, max_tokens, and temperature. Image specific options are passed alongside those: aspect_ratio takes one of the enum values listed above, n is fixed at 1, and input_references accepts between 0 and 1 reference images. If you have existing OpenAI client code, switching to this model is largely a matter of changing the model id and adding the image parameters to your request body.

The main constraints to weigh are the single output image per request and the single reference image limit. If your pipeline needs several images from one call, or needs to combine multiple references into one generation, you will have to issue multiple requests or look at a different model. The 4096 token context is also modest, so very long prompts or detailed reference descriptions may need trimming. For straightforward prompt driven generation with one anchor image and a chosen aspect ratio, it fits cleanly.